A hybrid InSAR–artificial intelligence/machine learning ensemble framework for evaluating the potential impact of a dam on land subsidence susceptibility

Study region Dehloran Plain. Study focus This study presents a scenario-based methodology to investigate the potential impact of the Meimeh storage-recharge dam, which is currently in the planning-to-construction phase, on downstream land subsidence susceptibility. A CHART-FLWD framework was developed and optimized using five well-known machine learning and artificial intelligence algorithms: ANN, XGBoost, RF, SVM, and GA, with model evaluation performed against InSAR-derived deformation observations representing the current pre-dam condition. Based on continuous evaluation metrics, GA (R² = 0.82), XGBoost (R² = 0.75), and ANN (R² = 0.70) showed stronger spatial agreement with the InSAR-derived deformation patterns, whereas RF (R² = 0.30) and SVM (R² = 0.40) showed lower continuous prediction performance. The optimized results indicated that groundwater recharge and drawdown are the primary drivers of subsidence susceptibility in the region. Accordingly, a new optimized ensemble LSS model, named GAX, was developed and applied to the current pre-dam condition and to a simulated post-dam scenario by incorporating dam-related variables, including proximity to rivers and the planned irrigation network. New hydrological insights The findings revealed a significant reduction in subsidence intensity, with index values decreasing from 39 to 110 before dam construction to 18–101 under the hypothetical post-dam scenario, highlighting the potential hydrological influence of the dam scenario. High and very-high risk areas declined from 68% to 31%, extreme-risk zones nearly disappeared, while moderate and low-risk areas expanded. Given the increasing subsidence severity, the proposed methodology provides a useful framework for identifying high-risk zones and supporting sustainable groundwater-management strategies. The proposed framework demonstrates potential applicability to hydrogeologically similar arid and semi-arid alluvial plains, although further multi-regional validation is required.